Data mining of meteorological-related attributes from smartphone data

• Autor(a) principal: Brito, Luiz Fernando Afra
• Data de Publicação: 2017
• Outros Autores: Albertini, Marcelo Keese
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: Repositório Institucional da UFLA
• Texto Completo: http://repositorio.ufla.br/jspui/handle/1/15025
• topic: Data mining; Rainfall; Smartphones; Signal strength

This paper presents studies on using data mining techniques with data collected from mobile devices in order to verify the viability of usage on rainfall alert systems.In our study, we have employed smartphones to gather meteorological-related data from telecommunication technologies, such as, Global System for Mobile Communications (GSM) and Global Positioning System (GPS). In order to evaluate the capability of monitoring rain with data from smartphones, we used a simplified rainfall simulator to conduct studies in controlled scenarios.We used classification algorithms such as k-Nearest Neighbors, Support Vector Machine and Decision Tree to identify rainfall types (no rain, light rain and heavy rain). The classification results were promising and showed area under ROC curve of 0.95 with the k-Nearest Neighbors algorithm and 0.80 with Support Vector Machine. Additionally we have conducted preliminary and promising experiments in a real world scenario which motivates further research on data collection, preprocessing and specialized classification for alarm systems.